跳到主要內容

Hermes Agent 開源自學 AI 助手:NousResearch 打造會成長的個人 Agent | Hermes Agent: NousResearch Open-Source AI Agent That Learns and Grows With You

By Kit 小克 | AI Tool Observer | 2026-04-09

🇹🇼 Hermes Agent 開源自學 AI 助手:NousResearch 打造會成長的個人 Agent

Hermes Agent 是什麼?開源 AI Agent 新標竿

Hermes Agent 是 NousResearch 推出的開源自學 AI Agent,核心理念是「跟你一起成長的 Agent」。不同於一般靜態的 AI 助手,Hermes Agent 會從你的互動中學習,自動建立技能、累積記憶,越用越懂你。2026 年 4 月 8 日剛發布 v0.8.0 版本,合併了 209 個 PR、解決 82 個 issue,是目前最完整的開源個人 Agent 方案之一。

Hermes Agent 有哪些核心功能?

Hermes Agent 最大特色是「自我進化」,不只回答問題,還能執行任務、管理工作流程。以下是主要功能:

  • 跨平台整合:支援 Telegram、Discord、Slack、WhatsApp、Signal、CLI,一個 Agent 管所有通訊
  • 自動學習迴圈:從對話中自動建立技能,使用時持續改進,並主動保存知識
  • 40+ 內建技能:涵蓋 MLOps、GitHub 工作流、研究輔助等常見場景
  • 六種終端後端:本地、Docker、SSH、Daytona、Singularity、Modal,支援無伺服器持久化
  • 即時切換模型:v0.8.0 新增 /model 指令,對話中途可切換 AI 模型和供應商
  • MCP OAuth 2.1:符合標準的 OAuth 認證,安全連接各種 MCP 工具伺服器

v0.8.0 版本帶來哪些重要更新?

這次「智慧版」更新聚焦在 Agent 的自主能力和安全性,是一次質的飛躍。

  • 背景任務自動通知:長時間執行的任務完成後自動提醒,不需手動輪詢
  • 自優化工具引導:系統自動診斷並修補了 GPT/Codex 模型的五種失敗模式
  • 智慧閒置超時:追蹤實際工具活動而非計時器,正在工作的任務不會被強制中斷
  • Google AI Studio 原生支援:Gemini 模型直接可用,自動偵測上下文長度
  • 安全強化:SSRF 防護、時序攻擊緩解、OSV 惡意軟體掃描

Hermes Agent 跟 Claude Code、ChatGPT 有什麼不同?

最大差異在於 Hermes Agent 是完全開源、可自架的個人 Agent,你的資料不會離開你的機器。

  • 開源自架:程式碼完全公開,可在自己的環境部署,隱私掌握在自己手中
  • 模型無關:不綁定單一 AI 供應商,可用 OpenAI、Google、Anthropic 或開源模型
  • 持久記憶:跨對話累積對你的理解,不像一般 chatbot 每次重新開始
  • 社群驅動:18 位核心貢獻者加社群共同開發,更新速度快且透明

實際使用體驗如何?

安裝後綁定你的通訊帳號,Hermes Agent 就會開始在背景運作。它會觀察你的工作模式,逐步建立專屬技能。例如你常用 GitHub,它會自動學會你的 PR review 流程。跨平台整合意味著你在 Discord 交代的事,可以在 Telegram 追蹤進度。

不過要注意,目前設定過程仍需要一定的技術門檻,適合有基礎開發經驗的使用者。社群文件正在快速完善中。

FAQ 常見問題

  • Q:Hermes Agent 免費嗎?
    A:完全開源免費,但使用的 AI 模型可能需要付費(如 OpenAI API),也可搭配免費開源模型。
  • Q:需要什麼硬體才能跑?
    A:基本需求不高,一般筆電即可運行 CLI 版本。若使用本地模型則需要 GPU。
  • Q:支援中文嗎?
    A:支援程度取決於你選用的底層模型,搭配 GPT-5.4 或 Gemini 可有良好中文體驗。

好不好用,試了才知道。


🇺🇸 Hermes Agent: NousResearch Open-Source AI Agent That Learns and Grows With You

What Is Hermes Agent? A New Open-Source AI Agent Benchmark

Hermes Agent is an open-source, self-improving AI agent built by NousResearch. Its core philosophy is simple: an agent that grows with you. Unlike static AI assistants, Hermes Agent learns from your interactions, automatically builds skills, accumulates memory, and gets better the more you use it. Version 0.8.0 dropped on April 8, 2026, merging 209 PRs and resolving 82 issues — making it one of the most complete open-source personal agent solutions available.

What Are the Core Features of Hermes Agent?

Hermes Agent stands out with its self-evolution capability — it does not just answer questions but executes tasks and manages workflows. Key features include:

  • Cross-platform integration: Works across Telegram, Discord, Slack, WhatsApp, Signal, and CLI from a single gateway
  • Auto-learning loop: Creates skills from conversations, improves them during use, and proactively persists knowledge
  • 40+ built-in skills: Covers MLOps, GitHub workflows, research assistance, and more
  • Six terminal backends: Local, Docker, SSH, Daytona, Singularity, and Modal with serverless persistence
  • Live model switching: The new /model command in v0.8.0 lets you change AI providers mid-conversation
  • MCP OAuth 2.1: Standards-compliant authentication for secure MCP tool server connections

What Does the v0.8.0 Release Bring?

This "intelligence release" focuses on agent autonomy and security, representing a major leap forward.

  • Background task notifications: Long-running tasks automatically alert the agent upon completion
  • Self-optimized tool guidance: The system diagnosed and patched five failure modes in GPT/Codex model behavior
  • Smart inactivity timeouts: Tracks actual tool activity rather than wall-clock time
  • Native Google AI Studio support: Gemini models work directly with automatic context length detection
  • Security hardening: SSRF protections, timing attack mitigations, and OSV malware scanning

How Does Hermes Agent Compare to Claude Code and ChatGPT?

The biggest difference is that Hermes Agent is fully open-source and self-hosted — your data never leaves your machine.

  • Self-hosted: Full code transparency, deploy in your own environment, complete data privacy
  • Model-agnostic: Not locked to any single AI provider — use OpenAI, Google, Anthropic, or open-source models
  • Persistent memory: Builds understanding of you across sessions, unlike chatbots that start fresh every time
  • Community-driven: 18 core contributors plus community, with rapid and transparent development cycles

What Is the Real-World Experience Like?

After installation, you connect your messaging accounts and Hermes Agent starts running in the background. It observes your work patterns and gradually builds personalized skills. For example, if you frequently work with GitHub, it will learn your PR review workflow. Cross-platform integration means you can assign a task on Discord and track progress on Telegram.

That said, the setup process still requires some technical knowledge. It is best suited for developers with basic command-line experience. Community documentation is improving rapidly.

FAQ

  • Q: Is Hermes Agent free?
    A: It is fully open-source and free. However, the underlying AI models may have costs (e.g., OpenAI API), though you can pair it with free open-source models.
  • Q: What hardware do I need?
    A: Basic requirements are modest — a standard laptop can run the CLI version. Local models require a GPU.
  • Q: Does it support multiple languages?
    A: Language support depends on your chosen base model. GPT-5.4 or Gemini provide excellent multilingual capabilities.

You will not know if it works until you try it.

Sources / 資料來源

常見問題 FAQ

Hermes Agent 是什麼?

Hermes Agent 是 NousResearch 開發的開源自學 AI Agent,會從使用者互動中學習並自動建立技能,支援 Telegram、Discord、Slack 等多平台。

Hermes Agent 免費嗎?

完全開源免費,但底層 AI 模型可能需要 API 費用,也可搭配免費開源模型使用。

Hermes Agent 跟 ChatGPT 有什麼不同?

Hermes Agent 是開源可自架的個人 Agent,資料不離開你的機器,支援多模型切換,且具備跨對話持久記憶。

Hermes Agent 需要什麼硬體?

基本 CLI 版本一般筆電即可運行,若要跑本地模型則需要 GPU 支援。

Hermes Agent v0.8.0 有哪些新功能?

主要新增即時模型切換、背景任務自動通知、Google AI Studio 原生支援、MCP OAuth 2.1 認證和多項安全強化。

延伸閱讀 / Related Articles


AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends

留言

這個網誌中的熱門文章

MCP 突破 9700 萬次下載:AI Agent 的「USB-C」為何成為 2026 年最重要的標準? | MCP Hits 97 Million Downloads: Why Model Context Protocol Became the Most Important AI Standard of 2026

歡迎來到 AI 工具觀察站 | Welcome to AI Tool Observer

GitHub Copilot 預設用你的程式碼訓練 AI:4 月 24 日前必須手動退出 | GitHub Copilot Will Train AI on Your Code by Default — Opt Out Before April 24